Comments on Advanced NFL Stats: Run-Pass Balance--A Historical Analysis

There will forever be intangibles, because human b...

2011-12-02T13:19:50.940-05:00

There will forever be intangibles, because human beings play the game. Football intelligence, body type, athleticism, preparedness, and emotional state on a particular day/play - lots of things influence execution. I have not seen any references to what happens after a play is called and before it has ended. Generalized statistics include results by highly effective and highly ineffective players and coaches - all blended together into an homogenized puree. This data seems to assume that all players are equal. The best play callers and the best play makers seem to be those who are able to adapt most effectively to what the other team is doing and to outperform them in the field. Tendencies really should only be applied to the play calling of particular coaches and the trends in execution of particular players.

Numerically it makes sense to pass on every play s...

2011-11-21T19:58:03.186-05:00

Numerically it makes sense to pass on every play since you are at an advantage on all plays by spreading the field and limiting the amount of available blitzes the OL need to pick up.

However this will likely result in smart defensive coordinators blitzing everyone in an attempt to remove the starting QB from the game. In fact this would be the most proficient way to play defense. Even if you give up a TD, injuring the QB by sending all 11 on the first pass play of the game would result in the opponent, particularly if the opponent has a star QB, scoring far fewer points than if yo played a conventional defense.

If i were coaching a team with a limited QB, i would make it know that more than 20 passes per game will result likely in QB injury.

Really interesting post here... Obviously the Mel...

2011-10-12T23:17:37.150-04:00

Really interesting post here...

Obviously the Mel Blount rules changed the passing game forever.

But there have been many small changes over the years that I think have led to offenses creating more points. People like points in general, few fans love a 3-0 game (I'm one).

However, not all rule changes are created equal. And some will have smaller steady impacts. For example the new rules regarding hitting defenseless receivers will probably lead to a few more caught passes and maybe even some more YAC when a tackle is missed that maý have killed the receiver before.

And all these small changes correlate into increased passing prowess as time goes on.

Passing teams tend to be more successful because t...

2011-05-08T15:19:08.177-04:00

Passing teams tend to be more successful because the QB are in control.

What also would be interesting to look at is if th...

2011-03-19T17:02:12.741-04:00

What also would be interesting to look at is if there's a major penalty differential in running vs. passing. For one, I imagine that, on average, there are more holding calls on pass plays. On the flip slide, do unnecessary roughness calls, pass interference, defensive holding calls account for any statistically significant benefit? This is just purely speculation, but isn't the general wisdom that NFL penalties are weighted to the pass game vs. run game?

To highlight this point, consider that the teams g...

2011-02-26T23:42:42.415-05:00

To highlight this point, consider that the teams generally known as "running teams" generally have the most effective defenses.

(The Chicago Bears in the 1980's, the New York Giants in the 1990's, the Baltimore Ravens in the early 2000's, and the Pittsburgh Steelers as examples)

Simply put, a defense that is built to be on the field for a below-average number of possessions per game can be built around athletes who have more "quick burst" and power as opposed to more endurance athletes required by passing teams (white muscle fiber vs red muscle fiber).

Running the ball has other advantages. First, run...

2011-02-26T23:36:07.351-05:00

Running the ball has other advantages.

First, running the ball effectively uses up the play clock, and generally "shortens the game" in terms of the total number of possessions during the game for each team.

A great example of this is where the New York Giants, chiefly a running team, defeated the Buffalo Bills, an aerial power, in Super Bowl XXV.

Another important point to make is that shortening the game by running the clock down also gives the defense a chance to rest while the offense is on the field.

Conversely, in this scenario, the opposing defense is not allowed a chance to rest, and is worn down by the running team's offensive line.

The numbers in aggregate are interesting, but not ...

2011-02-22T15:34:59.083-05:00

The numbers in aggregate are interesting, but not revelatory. Increasing short-yardage running situations will bring down the average yards per rushing attempt, but does not speak about the success of a running play. Yes, not every "running down" will include a run, but I imagine that runs face a steeper yardage penalty than a pass in those situations.

A more honest evaluation here is the expected YPC/YPA on 1st and 10, or an analysis of short-yardage running plays.

Is it possible that the substantial decline in Pas...

2010-12-03T15:40:25.503-05:00

Is it possible that the substantial decline in Pass YPA in the 1960s and 70s is due to a dramatic increase in the number of starting QBs? In 1959 there only 12 starting QBs. By 1978 there were 28 starting QBs. The main reason for this is the inclusion of 10 AFL teams plus a few more NFL expansion teams. As you add more teams the number of talented QBs does not automatically expand, so it just may be that many previous backups were suddenly thrust into starting positions.
A simple test would be to chart the Pass YPA for the top 12 QBs from 1959 to 1978 to see if this nullifies the downturn.

Yes they have “evened out” over the years because ...

2010-08-12T16:25:25.220-04:00

Yes they have “evened out” over the years because no coach has been silly enough to abandon the run in favor of the "all pass play" suggestions mentioned a few times above (Don Coryell comes to mind as someone who found this out the hard way, stats people always liked to mention how "bad" the Chargers’ Defenses were missing the simple point that the pass dominant offense used less time and kept the Chargers’ Defense on the field more thus fatiguing them and making them less effective.) I guess my response is fueled by this "analysis of stats in a vacuum" style thinking that permeates so many "stats" discussions and drives me crazy. The stats that "even out" due so because coaches "in the know" don't allow them to get out of hand by ignoring important real life factors like time and fatigue... If I'm missing your point completely I apologize I haven't analyzed your Win Probability model yet so I don't know how much of this you've factored in elsewhere...

Yes. The clock, score, and other aspects of the si...

2010-08-12T15:45:02.646-04:00

Yes. The clock, score, and other aspects of the situation is considered in my Win Probability model. Unfortunately, there isn't digital play by play data available to apply the model before 2000.

The clock and score certainly matter in any single situation, but over the course of an entire season or many seasons, those considerations even out. In other words, each season will feature roughly equivalent numbers of plays in various situations.

One more comment, and I realize this seems too eth...

2010-08-12T15:43:30.008-04:00

One more comment, and I realize this seems too ethereal a thing to put into a stat analysis but there is the factor of fatigue. Pass Blocking and Run Blocking are two completely different tasks that demand different levels of energy from the player. Similarly Run Defense and Pass Rush do too. If running the ball fatigues your opponents more (the corollary benefit of consuming time is keeping the opponent's defense on the field longer) it will have a corresponding effect on 3rd and 4th quarter effectiveness. On paper Ali never beats Forman but by using the fatigue factor Ali defeats Forman easily... How can fatigue be factored into the analysis of run vs pass on a statistical level?? Any suggestions? (p.s. - many wide receivers rest on run plays, even though they should be blocking... and running backs catch their breath pass blocking too... though how you'd factor this in is hard to say in a statistical world but in real football fatigue is a factor in how and why plays are chosen.)

Someone finally mentioned the clock! Shouldn'...

2010-08-12T14:56:12.684-04:00

Someone finally mentioned the clock! Shouldn't all this data be analyzed not just by how many points or yards runs or passes generate but how many points or yards they restrict from the opponents by consuming time!? It seems to me that that way a more balanced idea of the usefulness of Run vs Pass could be understood. Time is too important a factor to be left out of the equation as, unlike say Chess (my favorite game), it's not equal for both sides. Less time on the field = less chances to score and therefore less overall points generated. So a Run play would generate less over all points scored per attempt but more opponent's points restricted per play, or something like that... What do you think? (maybe you could find a mathematical expression of "the overall time of possession divided by scoring drives per game" and translate that into points restricted by running plays, or something, just thinking off the top of my head here...) Am I off base here!?

Yes. That is well understood, and that's what ...

2010-08-07T11:51:46.227-04:00

Yes. That is well understood, and that's what was meant by, "Without detailed play-by-play data, we can't rely on advanced metrics like Expected Points Added (EPA). Instead, simple efficiency will have to do."

Besides, those circumstances in which yardage is less important than time or other considerations will exist across all eras of the NFL.

There is more to playcalling than simply maximizin...

2010-08-07T11:48:18.959-04:00

There is more to playcalling than simply maximizing the yards per play average. Running plays specifically are useful in running down the clock, reducing the risk of turnovers, wearing down the defense, etc. So it makes sense for coaches to call these plays in many circumstances, knowing full well that they are sacrificing yardage for some other benefit.

I ran a few numbers on play calling tendencies and...

2010-06-16T15:38:47.411-04:00

I ran a few numbers on play calling tendencies and there are clear (and expected) patterns by quarter.

The numbers below are from the 2007 play-by-play data. It show the percentage of the time that a pass play (P) follows a pass, a rush (R), a penalty (Pen), or some other play (O) (such as kickoff or punt). The data are sorted by team and time, so it includes the start of a new series compared to the end of the last series (rather than comparing it to the last play of the other team. This also means that the end of one quarter in one game will be be followed by the beginning of the same quart the next week -- but accounting for that was too much work at this point.)

In a nutshell:
* Q1 & Q3: teams definitely tend to alternate
* Q2: there is no clear pattern
* Q4: teams tend to repeat the same play type.

Q1
__________O_____P___Pen_____R
P______13.47__34.97__5.61__45.94
R______16.06__41.51__5.63__36.80

Q2
__________O_____P___Pen_____R
P______10.93__45.09__5.48__38.50
R______14.84__43.98__5.26__35.92

Q3
__________O_____P___Pen_____R
P______12.79__38.52__5.40__43.30
R______15.79__43.11__4.63__36.48

Q4
__________O_____P___Pen_____R
P______10.92__52.20__5.58__31.30
R______11.42__37.81__4.76__46.01

Q5
__________O_____P___Pen_____R
P______14.29__36.51__4.76__44.44
R______13.13__26.26__4.04__56.57

Obviously there is more analysis that could be done --- looking at the data by team, by score, by yards to go ....

Pat-I've reviewed that article again. You'...

2010-01-30T14:38:22.639-05:00

Pat-I've reviewed that article again. You've quite certainly created a strawman, two of them actually. You arbitrarily constructed non-linear utility curves with absolutely no basis in empirical fact. You've literally dreamed them up.

Everyone agrees that the very premise of minimax theory is the existence of linear utility functions. In my model, as long as one considers every net point as equally valuable as another (2 points is twice as good as 1 point), then linearity holds. Further, a point scored while "learning" is equally valuable as one scored later.

As you may be aware, I was the one who pointed this out in my criticism of the Levitt-Kovash study. To overcome this problem, only situations where the score is close and time is not yet a factor are considered.

By the way, do cornerbacks get winded too?

Heh, guess I could have saved time then by clickin...

2010-01-27T23:56:19.699-05:00

Heh, guess I could have saved time then by clicking through (just got linked to this blog, I haven't read most of your work).

Pat-That's absurd. Players didn't get exha...

2010-01-27T22:47:52.144-05:00

Pat-That's absurd. Players didn't get exhausted in the 70s? Coaches didn't 'learn' in the 60s?

You are trying too hard.

Sunpar-Yes, those things can be accounted for. See...

2010-01-27T22:45:24.967-05:00

Sunpar-Yes, those things can be accounted for. See some of the other articles linked to above in the main post.

Hey Brian, Great post. I have just a few comments...

2010-01-27T22:16:59.991-05:00

Hey Brian,

Great post. I have just a few comments.

You are using yards per play as your measure of efficiency, but we all know that there is more to the efficiency of a play than just the number of yards it gains. Ultimately, the efficiency of a play is the amount to which it increases your win expectation. To that end, I feel that we need to compare runs vs. passes by looking at the following attributes of a play:

a. The potential for a turnover. I know that league interceptions have been going down, but so have fumble rates (read that from a reputable source). But is one going down more than the other? Your YPC doesn't include fumbles, while the YPP includes INT. Thus, it is likely that running efficiency is actually increasing a bit. For that matter, how have fumbles on passing plays changed? Are strip-sacks more or less prevalent? How about receiver fumbles?

b. The standard deviation of yards gained. Are running plays more predictable? We know that passing completion % has gone up recently, which suggests it's decreasing its standard deviation; but has running also become a more (or perhaps less) predictable play?

c. The last thing is somewhat esoteric and I have no concrete suggestions for how to test this, but I'll just put this out there: I've heard countless people talk about the value of "shortening" a game. That is, using the run to chew up clock and limit the amount of possessions and available time for the opposing team, thus constricting the other team's number of plays and, later in the game, their decision matrix. So you may say that with a certain lead, and at a certain point in the game, running is more efficient than the actual yardage metrics will show, becaue of the guarenteed running of clock. Obviously, you aren't going to go all run after you get a 14 point lead with 5:00 left in the 2nd quarter. But it maybe that it's in your advantage to skew the run:pass ratio as the game marches on, to shorten the opportunity for your opponent to come back.

Now, as for how the above could effect run/pass balance: what if the increase in pass efficiency is allowing teams to build sustainable leads EARLIER and MORE OFTEN than before? With bigger leads earlier in games, perhaps teams are shifting to a heavier run attack as game time passes.

1) I haven't been a kid in a very long time. ...

2010-01-27T21:02:32.128-05:00

1) I haven't been a kid in a very long time.

2) A strawman is an example based on a misunderstanding (or misrepresentation) of the original position. Your position is that "eventually, there is an equilibrium where the payoffs [for run and pass] should be equal." I gave several examples where this would not be the case. That's not a straw man. You might believe the examples are ridiculous, although I've given support for the basic idea, but that still wouldn't be a straw man.

The previous changes - in the late 70s and in the 60s - might not be comparable, because in both of those cases the trend was either towards more balanced playcalling (i.e. the *infrequent* action was more effective) or were only moderate (or short in time) deviations.

In this case, you've got the dominant action having an advantage for a long time, which implies that the game has a sort of "natural wall" to it.

That behavior *does* shows up in both the "progressive (learning) defense" model and the "player exhaustion" model - if you look here, the stable equilibrium there is exactly that kind of "wall".

More strawmen. You didn't answer the question,...

2010-01-26T20:11:56.297-05:00

More strawmen. You didn't answer the question, kid. Why no response?

Geh, missed the support of #3. That's supporte...

2010-01-26T19:41:17.125-05:00

Geh, missed the support of #3. That's supported by the fact that runs are the only plays where offensive linemen can immediately move downfield.

Also #1 would also be supported with fatigue/injury to the quarterback as well.

"You're off on a contrarian semantic tang...

2010-01-26T19:34:39.144-05:00

"You're off on a contrarian semantic tangent throwing up strawmen."

Which one's a strawman? Each possibility is supported by both game theory and actual evidence from football coaches when they explain playcalling.

Passing has been more successful than rushing since 1978, when the rules dramatically changed. That's 31 years.

If there was a more successful playcalling strategy available, it's highly unlikely no one would've found it in 31 years. They started to adapt to a newer, better strategy that had become available nearly immediately in 1978. The lag that you mentioned before 'modern' efficiency was hit was about 10 years. It's been twice that now.

It is far, far more likely at this point that there's some basic feature of the game that discourages more extreme levels of passing.

"Why do you think there hasn't been a response in the proportion of passing despite the recent long trend of increasing efficiency? There was a response after '78. There was a response after the decline in the 60s."

I've offered several possibilities.

1) Passing more simply isn't realistically possible due to fatigue. This has some intuitive support, since passing plays cover much larger distance than running plays.

2) Successful passing requires determining a defense's weakness, which is best done via safer plays. This is supported by statements from coaches.

3) Successful playcalling requires manipulating defensive player actions, and runs manipulate defenses in ways that other passes cannot.

It's important to state that it's hard to prove any of these statements, as they would be *beliefs* of coaches rather than facts. But I don't see why it's any less likely than the belief that teams would do better if they passed more.

There are other interesting possibilities as well. Coaches could be optimizing for something other than gain per play - with football being such a low-scoring game, it's possible that average results would just require too long of a game for the best strategy to win out.